The Oxford Handbook of Neolithic Europe

by Chris Fowler

  Clarification is now beginning to come from the gradual accumulation of aDNA (ancient DNA) analyses of prehistoric populations, but the interpretation of these also requires caution. Haak et al.’s (2005) study of ancient mtDNA from Linearbandkeramik (LBK) skeletons in central Europe found that the most frequent type, present in 25% of the samples, is extremely rare today, occurring in only 0.2% of the population—a decline in frequency that is very unlikely to be simply the result of genetic drift alone. The authors favoured the idea that this could be accounted for if a very small number of pioneer immigrant farmers were soon overwhelmed by local foragers adopting farming. They also noted, but largely discounted, the possibility that it could be explained by a subsequent population replacement. Either way, the inference is that the LBK farmers contributed little to the genetic composition of modern European populations.

  A more recent study comparing aDNA patterns for European Mesolithic hunter-gatherers and LBK early Neolithic farmers and modern mtDNA samples from European populations has radically changed this picture (Bramanti et al. 2009). It has confirmed the probable lack of continuity between LBK farmers and modern Europeans, but has also shown that there is no continuity between the hunter-gatherer and modern populations either. More importantly from the archaeological point of view, it appears that there is no genetic continuity between the Mesolithic and early Neolithic populations, confirming the demic diffusion model for the spread of farming, at least in this part of Europe. It is particularly striking that there is a lack of continuity in the mtDNA, indicating female descent, because many models that include a demic diffusion element still assume the incorporation of hunter-gatherer females into the farming population (e.g. Price et al. 2001).

  Another example of population extinction and replacement processes is provided by the results of Linderholm’s (2008) recent aDNA study of Neolithic populations in Sweden. Here a very clear archaeological distinction between the middle Neolithic Trichterbecher (TRB) culture and the contemporary Pitted Ware culture has long been recognized. Linderholm’s analysis demonstrated that these represented two genetically distinct populations, and that whereas the genetic patterns of the TRB population continued into the local Bronze Age, the Pitted Ware patterns appear to die out. Interestingly, the TRB population also showed a higher frequency of the lactose tolerance gene, potentially reflecting a history of stronger natural selection related to milk consumption in this population. A recent simulation study suggests that selection for the lactose tolerance gene started around 7,500 years ago in a region between the Balkans and central Europe (Itan et al. 2009); aDNA evidence indicates that frequencies were initially low (Burger et al. 2007), as would be expected.

  In summary, we should be very careful in making statements about the composition of the present-day population in terms of its ancestry at particular times and places during the Neolithic, or earlier, because we cannot assume that there was massive regional population continuity up to the present. Ancient DNA evidence is beginning to tell us what we want to know about populations in the past much more directly than the extended chains of assumption and inference from the present that previously were the only available option. The complex history of population extinctions and replacements that aDNA studies are now revealing is also suggested by alternative lines of evidence for regional population fluctuations (e.g. Shennan and Edinborough 2007, and see below).

  LANGUAGE

  Whilst the genetic evidence has increased enormously in quantity and quality since it was first brought into the debate about the origin and spread of the European Neolithic, the continuing debates about language have not seen a corresponding improvement in the quality of the evidence, and without the discovery of new ancient written sources they never will. Thus, language is virtually always, in effect, the dependent variable in discussion—its history being inferred and its distribution patterns explained on the basis of putative links with the evidence of archaeology and genetics, which is then correlated with the known linguistic patterns.

  Since Renfrew’s initial proposal that the spread of Indo-European languages resulted from the demic expansion of farming into Europe, the debate has continued to focus on this language family and mechanisms for its spread, with an ongoing dichotomy between those who take the traditional view of a late date for proto-Indo-European at around 6,000 years ago and those who take Renfrew’s view. This dichotomy has broad ramifications for views of what went on in the European Neolithic and Bronze Age.

  For the traditional Indo-Europeanists it is not just a matter of language distributions but of a set of social institutions that are characteristically ‘Indo-European’. Whilst some authors (e.g. Anthony 2008) continue to subscribe to the kurgan hypothesis and see both languages and social institutions as the cultural baggage associated with expanding populations of steppe origins, others (e.g. Kristiansen 2005) take a more differentiated view, suggesting that the institutions, which, it is claimed, are evidenced in the archaeological record, and by inference the languages, which are not, spread through a combination of more local migration and a process of ‘peer polity interaction’ (Renfrew 1986) linked to the spread of prestigious innovations and increased mobility. Unfortunately, we have no way of knowing whether the proposed Indo-European institutions of the late Neolithic and early Bronze Age were exclusively associated with Indo-European speakers, even less with people becoming Indo-European language speakers. On Renfrew’s hypothesis many of them would long have been Indo-European speakers anyway (Renfrew 2005). The unresolved issue here is whether the claimed Indo-European institutions are similar by virtue of common descent, shared with language; whether they spread ‘horizontally’ across linguistic boundaries, as social and economic innovations often do, perhaps leading people to drop their pre-existing language and start speaking a local Indo-European one for prestige reasons; or whether they arose as a result of convergent development (cf. also Robb 2009).

  In any case, there is increasing evidence that Renfrew’s argument for the early date of proto-Indo-European is correct. This is based on the application of phylogenetic analyses—initially developed for the construction of family trees based on genetic data—to linguistic data, in the form of lists of cognate or non-cognate words of similar meanings in different Indo-European languages, including ancient ones. In this case the specific purpose of the analysis is to estimate the most probable date for proto-Indo-European, the most recent common ancestor of the Indo-European languages. The methods are mathematically extremely complex and it is unnecessary to go into detail about them here, but they have been successively refined since their initial application, to take into account problems that have been raised (Gray and Atkinson 2003; Ryder and Nicholls 2010). The methods have consistently produced dates in the range 6000–7500 BC for proto-Indo-European, with the most recent study producing a 95% probability interval in the range 5100–7800 BC, centred at 6400 BC. After investigating the potential impact of a range of factors that might bias their results (e.g. failing to recognize ancient word borrowings), Ryder and Nicholls conclude that the only factor which could make a major difference to their results, and bring them more into line with the traditional date, would be if there had been a major change in the rate of word diversification in a coordinated fashion across all the Indo-European languages then in existence between 3,000 and 5,000 years ago. Of course, this is not impossible, but given that it would involve coordinated changes all the way from India to the western half of Europe it seems unlikely.

  In contrast to most existing views, Heggarty et al. (2010) propose that in Europe at least, proto-Indo-European did not break up initially into the distinct sub-families of Italo-Celtic, Germanic, and Balto-Slavic, which were late developments, but was characterized by a pattern of ‘dialect continua’. That is to say, the original uniformity of proto-Indo-European did gradually degrade as a result of local linguistic innovations taking place in particular places and not spreading throughout the whole area, but different innovations had diff
erent and often overlapping distributions, with specific innovations tending to be shared by places that were geographically closer. Thus, the reason that the Germanic sub-family shares features with both the Italo-Celtic and the Baltic-Slavic sub-families, and is therefore problematic for those constructing Indo-European language trees, is simply because of its geographical position between the two in central and northern Europe, and its consequent sharing of innovations from both sides. For Heggarty, the presence of such dialect continua is more in keeping with versions of Renfrew’s proposal that proto-Indo-European spread with the ongoing population expansion by small-scale societies that was associated with the spread of farming, rather than with the expansion of mobile warlike hierarchical societies 3,000 years later. However, in his view, it probably was the appearance of such societies that led to the emergence of a more clearly differentiated linguistic picture later on, as some initially small-scale societies expanded at the expense of others and their dialects (in the way that we know Latin, for example, expanded at the expense of other local Italic languages), and as local population movements led to the juxtaposition of groups who spoke different languages because of their previous spatial separation. In other words, there is not necessarily complete incompatibility between an earlier proto-Indo-European spread and some of the claims made by supporters of the more traditional view.

  Whether or not the varying degrees of demic diffusion associated with the spread of farming that are now accepted by the majority of authors were linked with the spread specifically of proto-Indo-European, we can still ask what its linguistic consequences might have been. If the lack of genetic continuity between incoming LBK farmers and local Mesolithic populations is also true for other parts of Europe, then the answer is language replacement. But the result is likely to be the same even if small numbers of foragers were successively incorporated into an expanding farming population, because the foragers would always have been coming in as a small minority; thus they, or at least their offspring, would be under strong pressure to adopt the farming language and the gradually changing language could potentially extend over far wider areas than the genes of ancestral farmers—this is Renfrew’s (2002) ‘Staged Population Interaction Wave of Advance’ model.

  However, it is not the only possibility. Ackland et al. (2007) developed an alternative model of the kinds of process that might have occurred. Once again, they assume demic diffusion arising from agriculture as the more productive subsistence strategy, and non-adaptive cultural features like language are simply carried along with the expanding population. The model includes a farming population with its cultural baggage and a forager population at much lower densities with its own traits, which converts from foraging to farming, its population potentially growing to the same level as its neighbours. In a geographically differentiated landscape, if the expanding population of farmers arrives at a boundary where, for example, soil conditions are less suitable, the front stops until the population behind the front has reached a sufficient level to make it worth expanding again. However, the new subsistence mode can still potentially spread to the forager population in advance of the wave, as for them it may be worthwhile because it gives them better returns than hunting and gathering. In these circumstances, if the main farming frontier is static for long enough, the forager population can grow sufficiently to provide a barrier to the expanding original farming population (even if that had incorporated a lot of indigenous foragers on the way). The result will be a cultural/linguistic boundary between the two populations.

  The fact that farming did not spread at a uniform rate, but was characterized in different regions at different times by periods of expansion and stasis, is now well established (e.g. Guilaine and Manen 2007; Bocquet-Appel et al. 2009). Thus, the potential for the sort of process modelled by Ackland et al. certainly existed, though whether it actually happened anywhere is uncertain (see below). That the process involved stops and starts also brings out the problems with the initial Ammerman and Cavalli-Sforza model of the spread of farming. Whilst it remains broadly true that the rate of spread they identified represents a valid first-order approximation to the overall rate of spread from the Near East to north-west Europe, they argued that the rate in itself gave an indication of the explanatory mechanism, demic diffusion by the wave of advance. This is because of the constancy of the rate this model produces and because the application of the ethnographically derived population growth rates and marriage distances to the diffusion equation gave the rate they found. Since the process was actually one of stop-and-go, with very different expansion rates at different times–as Ammerman and Cavalli-Sforza themselves acknowledged–this argument cannot be considered valid in itself as a justification for the demic diffusion model, even though, as we have seen, there are other arguments that strongly support it.

  These shortcomings have recently been addressed by Davison et al. (2006), who keep the same basic population dynamics and diffusion ‘wave of advance’ model but add a parameter to allow for the possibility that movement may not be equal in all directions; certain directions, for example along major rivers or coastlines, may be preferred. Moreover, local carrying capacities can vary, dependent, for example, on altitude and latitude. When the equations are run with the values that Davison et al. propose and justify for variations in the preferred direction of movement and carrying capacity, the fit of the model to the pattern of dates for the arrival of farming in different parts of Europe is very good, so the results do not just produce a description of the rate of spread but support demic diffusion as the mechanism after all.

  DEMOGRAPHIC PATTERNS

  As we have seen, virtually all the arguments above, about both genes and languages, depend on claims about population sizes and distributions. What evidence do we have for these during the periods in question?

  Gamble et al.’s (2005) study of population fluctuations in the western half of Europe in the late Palaeolithic and Mesolithic, based on summed radiocarbon date probability distributions as a population proxy, suggested that, with the exception of certain coastal areas, populations in the later Mesolithic were at historically low levels, presumably because the developing forest cover resulted in decreasing animal population densities. Figure 7.1 shows the summed date probabilities approach taken forward into the Neolithic for a number of broad areas where good data are available (see also Shennan 2009). In the majority of cases, very low late Mesolithic population levels are succeeded by massively increased levels with the initial arrival of farming. In Belgium, the Netherlands, and Germany this occurs with the arrival of the LBK in the second half of the sixth millennium BC. In England and Scotland we see the same phenomenon c. 4000 BC. Denmark and Ireland appear to be exceptions, with relatively high late Mesolithic population levels, though both show indications of a slight dip at the point where agricultural economies arrive.

  FIG. 7.1. Summed radiocarbon date probabilities taken as a proxy measure for the Neolithic populations of several regions of north-west Europe (from Weninger et al. 2009).

  Whilst care must be taken with the use of summed radiocarbon probability distributions as population proxies, the pattern of a decline in population levels over the course of the Mesolithic has also been recently shown at a more detailed level by Vanmontfort’s (2008) study of trends in Mesolithic occupation in several areas of the Low Countries on the basis of the chronological and spatial distribution of microliths. It appears that those specific areas of the Low Countries which subsequently became LBK early farming nuclei had long been devoid of Mesolithic occupation. Similar arguments have been made for a discontinuity or hiatus between late Mesolithic and early Neolithic in many parts of south-east and Mediterranean Europe, on the basis of gaps in site stratigraphies, especially caves (e.g. Berger and Guilaine 2009; Bonsall et al. 2001).

  It would appear, then, that the arrival of farming had a massive demographic impact in many areas, as predicted by the demic diffusion model. If an originally Near Eastern genetic composition
of the farming population had been successively diluted by the absorption of small numbers of foragers, as some versions of the demic diffusion model propose, then this might not have made much difference to the local genetic composition of the population, but at least with the LBK the results of Bramanti et al. (2009) are against this scenario. Even if it were valid in some areas, for the reasons explained in Renfrew’s ‘Staged Population Interaction Wave of Advance’ model, demic expansion would certainly have changed the languages spoken.